IRT Item Probability & Information Calculator
Enter calibrated IRT item parameters to evaluate the probability of a correct response, the item information at a selected ability, and the ability needed to reach a target probability.
Probability of a correct response60%
- Probability above the guessing floor
- 50%
- Item information at theta
- 0.48
- Single-item conditional SEM
- 1.44
- Ability needed for target probability
- 0.65
- Target ability relative to difficulty
- 0.65
- Effective discrimination used
- 1
- Effective guessing parameter used
- 20%
For 1PL, the calculator uses a = 1 and c = 0. For 2PL, it uses c = 0. For 3PL, it uses the entered c value.
Ability, difficulty, discrimination, and the scaling constant must come from the same calibration metric.
How to use this calculator
- Choose the 1PL, 2PL, or 3PL model used to calibrate the item.
- Enter ability, difficulty, discrimination, guessing, and the logistic scaling convention on one common metric.
- Set the target probability correct for the reverse ability calculation.
- Read the probability, information, conditional SEM, and target ability results.
How the item probability is calculated
This calculator evaluates a dichotomous item response theory item after its parameters have already been calibrated. Let theta be person ability, b be item difficulty, a be item discrimination, D be the logistic scaling constant, and c be the lower guessing asymptote. For the 1PL or Rasch model, the calculator uses a = 1 and c = 0. For the 2PL model it uses the entered discrimination and c = 0. For the 3PL model it uses both the entered discrimination and the entered guessing parameter.
The logistic core is L = 1 / (1 + exp(-D * a * (theta - b))). The probability correct is then P = c + (1 - c) * L. The value called probability above the guessing floor is (P - c) / (1 - c), which is the same as L after accounting for the lower asymptote.
How item information and SEM are calculated
The item-response derivative is Pprime = D * a * (P - c) * (1 - P) / (1 - c). Item information at the entered ability is I = Pprime^2 / (P * (1 - P)). Written without the derivative, this is I = D^2 * a^2 * ((P - c) / (1 - c))^2 * ((1 - P) / P). The single-item conditional standard error is 1 / sqrt(I). When information is effectively zero, the calculator reports the SEM as unbounded rather than dividing by zero.
What moves the result
Difficulty shifts the curve along the ability scale. Discrimination changes how steeply probability rises around the item's most informative region. A nonzero guessing parameter lifts the lower end of the curve, so at theta = b a 3PL item has P = c + (1 - c) / 2, not exactly 50%. The scaling constant also changes the slope, so parameters calibrated with D = 1.7 should not be mixed with a D = 1.0 calculation unless they were transformed.
Reverse-solving ability
For a target probability T, the inverted item characteristic equation is thetaT = b + ln((T - c) / (1 - T)) / (D * a). This is defined only when c < T < 1. Targets near the lower asymptote or near 100% can imply very low or very high abilities.
What this calculator leaves out
The calculator does not estimate item parameters from response data, test model fit, check dimensionality, inspect local independence, or evaluate differential item functioning. For test-level precision, item information should be summed across the administered items before converting to a conditional SEM.
Worked example
Suppose a 3PL item has theta = 0, b = 0, a = 1, c = 0.20, and D = 1.7. The logistic core is 0.5000, so the probability correct is 0.20 + 0.80 * 0.50 = 0.60, or 60.00%.
The derivative is 1.7 * (0.60 - 0.20) * (1 - 0.60) / 0.80 = 0.34. Item information is 0.34^2 / (0.60 * 0.40) = 0.4817, giving a single-item conditional SEM of 1.4409. A target probability of 80% requires ln((0.80 - 0.20) / 0.20) / 1.7 = 0.6462 logits above the item difficulty.
Common questions
What do the IRT difficulty, discrimination, and guessing parameters mean?
Difficulty locates the item on the ability scale. Discrimination controls how sharply probability changes as ability moves past that location. The guessing parameter is the modeled lower asymptote for a 3PL item, not a direct count of answer choices.
What is the difference between the Rasch, 2PL, and 3PL models?
The Rasch or 1PL model fixes discrimination and has no guessing parameter. The 2PL model estimates item discrimination but still has no lower asymptote. The 3PL model estimates discrimination and adds a guessing parameter.
Why is probability correct not 50% when ability equals difficulty in the 3PL model?
In a 3PL model, difficulty is the midpoint of the logistic part above the guessing floor. When ability equals difficulty, the probability is c plus half of the remaining distance to 1. With c = 0.20, that midpoint is 60%, not 50%.
Should I use a logistic scaling constant of 1.0 or 1.7?
Use the same convention used when the item parameters were calibrated. A value of 1.7 is often used to approximate the normal-ogive metric, while 1.0 is the plain logistic metric. Mixing conventions changes the slope and the reverse-solved ability.
What does item information say about where a question measures most precisely?
Item information is higher where a small change in ability produces a clear change in response probability. Higher information corresponds to a smaller conditional SEM. A single item is diagnostic; operational precision usually depends on the sum of information across items.
Can this calculator estimate IRT parameters from student responses?
No. It evaluates already calibrated parameters for one item. Estimating IRT parameters requires response data, model-fitting software, and checks that the chosen model is appropriate for the assessment.